Machine Learning Study of PFAS Intermediate Adsorption on Transition Metals: Scaling Relationships for Environmental Catalyst Design

Per- and polyfluoroalkyl substances (PFASs), particularly trifluoroacetic acid (TFA), have emerged as global environmental pollutants due to their extreme persistence. Conventional treatment methods are largely ineffective, underscoring the need for advanced catalyst design strategies. In this study, we develop a theoretical framework that integrates density functional theory (DFT) and machine learning (ML) to accelerate the discovery of PFAS degradation catalysts. By systematically evaluating the formation energies of 18 key degradation intermediates on 12 transition metal surfaces, we reveal consistent thermodynamic trends and uncover linear scaling relationships governing defluorination and carbon chain scission. These correlations enable dimensionality reduction of complex degradation networks, allowing pathway reconstruction using representative intermediates. Additionally, we introduce a machine learning model built upon knowledge-driven features capturing metal-adsorbate electronic interactions, functional group effects, and molecular property differences, which achieved an F1 score of 0.90 in classifying adsorption correlation patterns. Although the current framework is limited to this domain, it can be retrained with expanded data sets to cover broader PFAS families or additional catalysts. Together, our DFT-ML framework provides mechanistic insights and descriptor-guided strategies for PFAS degradation catalyst screening with potential applicability to other persistent organic pollutants.